Speculations about TSMC building a factory or acquiring a company in the United States

Yesterday, at TSMC's shareholders meeting, Chairman Liu Deyin said: "The company has evaluated the possibility of building a factory in the United States. The attitude is that it does not rule out the possibility and is also under consideration. However, due to the high cost of building a factory in the United States, a cost-effective solution has not been found, so there are currently no plans. At the same time, TSMC will also consider directly acquiring semiconductor factories for its layout in the United States, but they are all in the evaluation stage." He reiterated that TSMC's principle is to continue investing in Taiwan, and independent research and development will also be mainly based in Taiwan.

Liu Deyin also emphasized that he will personally attend the "Choose USA" investment summit. TSMC sends people to participate every year. This year he will attend in person. His purpose of attending is to take a look and express his views on the trade war. This attendance has nothing to do with Huawei.

As the world's largest wafer foundry, the voices of TSMC's senior management have always been the focus of the industry, and when it comes to mergers and acquisitions, they can attract even more attention. This time, on the topic of acquisitions, it is the second time that TSMC's senior management has expressed their views in recent times. The last time was in April this year, when the first quarter financial report was released. In response to some rumors, TSMC CEO Wei Zhejia said: "There are currently no acquisition plans. Of course, if there is a good opportunity that is in line with the company's strategy, we will consider it. At this stage, we will efficiently integrate existing assets to achieve further development."

At that time, GlobalFoundries made a series of moves: in February, GlobalFoundries sold its Singapore-based wafer factory Fab 3E to Vanguard International Semiconductor (VIS), a subsidiary of TSMC, for US$236 million; in April, GlobalFoundries announced that it had reached an agreement with ON Semiconductor to sell GlobalFoundries' 12-inch wafer factory Fab 10 in East Fishkill, New York, to the latter for US$430 million.

All these have aroused people's associations with TSMC's mergers and acquisitions. This time, the speculation about the company's mergers and acquisitions and factory construction has been further refined to the United States, adding another level of heat to the hot topic.

Although TSMC's production and investment base has always been in Taiwan, due to the increasing market activity in other regions (mainly mainland China) and customer concentration, TSMC's investment, factory construction and mergers and acquisitions outside Taiwan have always been topics of concern in the industry.

The biggest reason for building a factory in Nanjing, mainland China is naturally the huge and rapidly growing market; and the biggest reason for building a factory in the United States is customer concentration. According to statistics, TSMC's largest customer is still in the United States, which accounts for 61% of its revenue. Large companies such as Qualcomm, Broadcom, Nvidia, and Xilinx are its absolute major customers.

Being close to customers is undoubtedly the best choice for all manufacturing companies. However, building a factory in the United States traditionally faces a series of difficulties: First, recruiting people is a problem, and the labor cost is not low. Generally speaking, the labor cost of a wafer foundry accounts for more than 10% of the total investment. Moreover, as a core link in the supply chain, the logistics cost of IC manufacturing in the United States is not low.

But in recent years, the situation seems to be changing. First, Trump has been calling for the revitalization of the US manufacturing industry and has formulated a series of preferential policies for this purpose, the most representative of which is tax cuts, which is very attractive for building factories in the United States. Foxconn's Terry Gou made a bold attempt and began to build a panel factory in the United States, which has a certain demonstration effect.

In addition, Taiwan is a small island with limited power and water supply. However, the demand for electricity in wafer foundries is staggering. As TSMC expands production in Taiwan, its power consumption is increasing, while the supply capacity is becoming increasingly tight. This is also prompting the company to focus its future investment and expansion plans on regions outside Taiwan, with the main choices being mainland China and the United States.

As Liu Deyin said: Since the cost of building a factory in the United States is very high, a cost-effective solution has not been found. At the same time, TSMC will also consider directly acquiring semiconductor factories for its layout in the United States.

For a market like the United States, where manufacturing technology and factories are advanced but factory construction and labor costs are high, direct investment and acquisition seem to be more cost-effective than building a factory from scratch.

However, it is not easy for TSMC to find a suitable IC manufacturing target in the United States. First of all, the total number of IC manufacturing plants remaining in the United States is relatively limited, and they are all the core and most advanced IC businesses of major IDMs, so the probability of selling them is relatively low; in addition, TSMC's overall development strategy is to lead the most advanced process technology, and advanced processes are mostly used in the manufacture of logic and digital ICs, while IC manufacturing in the United States is mostly high-performance analog products, such as TI, ADI, ON Semiconductor, Qorvo and other IDMs, which are all analog IC manufacturers. Therefore, this seems to conflict with TSMC's strategy of developing the most advanced process.

In addition to IDM, the only companies with IC manufacturing business are wafer foundries. However, there seems to be only one foundry with manufacturing business in the United States. This is also the industry rumor mentioned earlier about TSMC's acquisition of a wafer foundry in the United States, and TSMC executives have denied this rumor.

It can be seen that whether it is building a factory or directly acquiring a target, there are too many factors to consider and the difficulty is not small.

The figure below shows the ranking of the world's top 15 semiconductor manufacturers in 2018 given by IC Insights.

Image source: IC Insights

As can be seen from the figure, among the top 15 semiconductor manufacturers, 7 are from the United States, and the others are scattered in Europe, Japan and South Korea. Among the 7 American manufacturers, 3 are fabless companies, namely Qualcomm, Broadcom and Nvidia, and these 3 are TSMC's major customers, which also reflects the necessity of building or purchasing factories in the United States.

Among the other four manufacturers, except for the analog giant TI, Intel, Micron and Western Digital have one thing in common, that is, memory chips are either their main business or an important business segment. The memory shortage and price increase in the industry in 2017 and 2018 made these manufacturers earn a lot of money.

So, would it be a good choice for TSMC to enter the memory business, especially in the United States?

In fact, in September 2018, at the SEMICON Taiwan, TSMC Chairman Liu Deyin’s statement caused quite a stir in the industry. It is reported that in his 20-minute English speech, he mentioned “storage” 14 times in total. He repeatedly emphasized that more than 80% of the energy consumed by current artificial intelligence computing is consumed by memory, which is a major bottleneck of current semiconductor technology.

Immediately afterwards, in an interview with the media, Liu Deyin admitted that TSMC "does not rule out acquiring a memory chip company", but there is no clear target at present.

At that time, both DRAM and NAND Flash would be within its consideration, because in the Toshiba TMC business acquisition case that made a lot of noise in 2017, TSMC had considered participating in the bidding to enter the NAND Flash market, but gave up because it failed to fully evaluate it. Last year, Liu Deyin's statement seemed to be more inclined to DRAM companies, and it was speculated that Nanya Technology in Taiwan was the main target.

In addition to its basic attribute of being a "commodity" in the field of integrated circuits, memory has also seen a demand for higher capacity and stronger performance as technology and applications develop. The simplest example is mobile phones, whose processor performance is getting stronger and stronger, and the requirements for DRAM are also getting higher and higher. At present, 6G and 8G memory are already common in flagship mobile phones, and it will not be long before 10G and higher capacity memory will enter more and more mobile phones.

In addition, AI is in its infancy, and AI itself is a very complex entity, which is very different from traditional processors such as CPU. AI aims to allow more human learning, thinking, and reasoning methods and logic to enter the machine. The training and reasoning processes in machine learning are becoming more and more complex, the algorithms are becoming more and more complex, and the amount of data involved is also increasing, all of which require the support of large-capacity, high-performance memory.

Therefore, traditional memory, especially DRAM, is increasingly unable to meet the application requirements of AI. People are already seeking technological breakthroughs to increase access capacity while ensuring indicators such as the speed of accessing instructions and data. In these areas, the industry has introduced a series of new storage technologies. Whether it is DRAM or NAND Flash, there are innovations every year, but it will take some time for them to be truly implemented.

For example, in AI applications, in order to cope with increasingly complex computing and processing requirements, the company has proposed the idea of ​​combining storage and AI processors into one, that is, integrating AI-specific chips into the processor to meet actual needs. In this way, the weight of memory, especially DRAM, in the entire system is constantly increasing. In traditional consciousness, whether it is a PC or a mobile phone, memory is only used to cooperate with the CPU or AP, and it is a supporting role. With the maturity of AI, this situation seems to be changing. In the future, it may be difficult for us to distinguish whether the AI ​​processor is important or the memory is important, because the actual application has higher and higher requirements for storage, and storage and processing seem to be moving towards integration.

Under such circumstances, it is natural for TSMC, which has deep technical accumulation in the semiconductor industry, as well as a strong customer and talent pool, to enter the memory business, and it is also the first choice for giant companies. With its semiconductor manufacturing level and capabilities, as well as its technical foundation and talent gathering ability, it is not difficult for TSMC to make a difference in the memory field.

Therefore, whether from the close relationship between AI development and storage, the proximity to the three major American customers, or the advantages of local memory research and development and manufacturing technology in the United States, the storage business is a good choice if you want to build or buy a factory in the United States.

Everything is just speculation, and the conclusion is subject to official information.